基于Caffe深度学习框架的车牌数字字符识别算法研究

Research of Recognition of Digital Characters on Vehicle License Based on Caffe Deep Learning Framework

摘要:

在车牌字符识别的某些场合中，获得的字符通常存在切割不均匀、光照对比度强烈、遮挡严重等强噪声污染。针对被强噪声污染的数字字符，本文提出基于Caffe深度学习框架的字符识别算法，在Caffe框架下搭建卷积神经网络，并对网络参数训练获得了一个鲁棒性强、识别精度高的网络结构。实验结果表明，在低噪声、中度噪声、强噪声污染情况下，文章中提出的方法相比当前典型的识别方法，在数字字符识别上均具有较好的识别能力，平均识别率高出将近5%，而在强噪声污染情况下，识别效果具有更加明显的优势。

Abstract:

In some license plate character recognition occasions, strong noises such as uneven cutting，strong illumination contrast and occlusion are inevitable. To solve this problem effectively, a character recognition algorithm based on Caffe deep learning framework was proposed in this paper. The convolutional neural network was built in the Caffe framework, and the network parameters were trained to obtain a network structure with high accuracy and robustness. Experiment results showed that the proposed algorithm has obvious advantages, the average recognition rate was about 5% higher when compared with the traditional digital character recognition method. And when there are strong noise pollution exists, the recognition results are better